regression metrics ignore 'compute=True' if 'multioutput="raw_values"

[jameslamb]

What happened:

Using the regression metrics like dask_ml.metrics.regression.mean_squared_error, I expected that passing y_true and y_pred with shape (n_samples, n_outputs) and multoutput="raw_values" would return:

• a Dask Array with shape (n_outputs,), if compute=False
• a numpy array with shape (n_outputs,), if compute=True

What you expected to happen:

Using dask_ml.metrics.regression.mean_squared_error() or dask_ml.merics.regression.mean_absolute_error() with multioutput="raw_values" always returns a Dask Array.

Minimal Complete Verifiable Example:

import dask.array as da
from dask_ml.metrics.regression import mean_squared_error

a = da.random.uniform(size=(100, 3))
b = da.random.uniform(size=(100, 3))

raw_output = mean_squared_error(a, b, multioutput="raw_values", compute=True)
type(raw_output)

# dask.array.core.Array

Anything else we need to know?:

I started looking into this while working on #756

Environment:

• Dask version:

output of 'conda list | grep -E "dask|distributed"'

dask                      2021.3.0           pyhd3eb1b0_0  
dask-core                 2021.3.0           pyhd3eb1b0_0  
dask-glm                  0.2.0                    pypi_0    pypi
dask-saturn               0.2.2                    pypi_0    pypi
dask-sphinx-theme         1.3.5                    pypi_0    pypi
distributed               2021.3.0         py37h06a4308_0  

• Python version, Operating System

output of 'conda info'

     active environment : None
       user config file : /home/jlamb/.condarc
 populated config files : 
          conda version : 4.9.2
    conda-build version : not installed
         python version : 3.7.6.final.0
       virtual packages : __glibc=2.27=0
                          __unix=0=0
                          __archspec=1=x86_64
       base environment : /home/jlamb/miniconda3  (writable)
           channel URLs : https://repo.anaconda.com/pkgs/main/linux-64
                          https://repo.anaconda.com/pkgs/main/noarch
                          https://repo.anaconda.com/pkgs/r/linux-64
                          https://repo.anaconda.com/pkgs/r/noarch
          package cache : /home/jlamb/miniconda3/pkgs
                          /home/jlamb/.conda/pkgs
       envs directories : /home/jlamb/miniconda3/envs
                          /home/jlamb/.conda/envs
               platform : linux-64
             user-agent : conda/4.9.2 requests/2.22.0 CPython/3.7.6 Linux/5.4.0-70-generic ubuntu/18.04.4 glibc/2.27
                UID:GID : 1000:1000
             netrc file : None
           offline mode : False

• Install method (conda, pip, source): installed from source as of f5e5bb4, using python setup.py install.